Lesion-Gated Hybrid Synthesis for Virtual Contrast-Enhanced Breast MRI: A MAMA-SYNTH Challenge Solution

Shohei Yoshimoto

eess.IVSep 18, 2026

Lesion-Gated Hybrid Synthesis for Virtual Contrast-Enhanced Breast MRI: A MAMA-SYNTH Challenge Solution

Authors: Shohei Yoshimoto

Organizations: Department of Radiology, The Jikei University School of Medicine Tokyo, Japan

Abstract

Purpose: Contrast-enhanced breast MRI depends on intravenous gadolinium-based contrast agents, motivating methods that synthesise post-contrast appearance from pre-contrast images alone. The MAMA-SYNTH Challenge (MICCAI 2026 Deep-Breath Workshop) evaluates such synthesis across four metric groups: image fidelity, tumour region of interest, downstream classification, and downstream segmentation. The four group ranks are averaged, so optimising a single objective is insufficient. Materials and Methods: We developed a lesion-gated hybrid synthesis pipeline. A lesion probability map, estimated from the pre-contrast slice alone by an ensemble of pre-contrast-only segmentation networks, spatially coordinates a tumour-focused regression pathway and a background-focused Pix2PixHD synthesis pathway, followed by a region-dependent calibration of the predicted enhancement. Training used the public MAMA-MIA collection with a patient-level split. Inference consumes a single pre-contrast 2D slice, with no mask, no post-contrast image, and no auxiliary metadata beyond image geometry. Results: On internal validation (n = 120 patients) the method reached MSE 0.523, LPIPS 0.185, and tumour-region SSIM 0.495. It was submitted to the hidden external 300-case test cohort as a self-contained inference container. The method ranked sixth in the official MAMA-SYNTH Challenge leaderboard. Conclusion: A test-compatible lesion probability map derived from the pre-contrast image can coordinate complementary tumour-focused regression and background-focused perceptual synthesis, enabling balanced virtual contrast enhancement under a multi-metric challenge setting.

Figures & tables

Item	Value
Task	Pre-contrast T1 fat-saturated 2D slice $\rightarrow$ peak-enhancement post-contrast 2D slice
Inference input	One 2D float32 z-score image; no mask, no auxiliary metadata beyond image geometry, no 3D context
Inference output	One image, identical shape; spacing, origin, direction preserved
Submission	Algorithm container; test phase limited to one submission
Cohort	MAMA-MIA, 1,506 patients; DUKE 291 / ISPY1 171 / ISPY2 980 / NACT 64; $>$ 25 US centres
Scanners	GE 966 / Siemens 411 / Philips 129

TABLE I: Challenge and data setup.

Fig. 1: Illustrative schematic of the lesion-gated hybrid synthesis pipeline. A lesion probability map is estimated from the pre-contrast slice alone and drives two gates: a hard dilated gate through which the lesion-focused regression pathway supplies enhancement, and a complementary soft gate through which the Pix2PixHD-based pathway supplies background texture. A final region-dependent calibration attenuates the predicted background enhancement while leaving tumour enhancement unchanged. Inference requires no mask, no auxiliary metadata beyond image geometry, and no 3D context. Component names in the figure are descriptive rather than architectural identifiers; exact configurations are given in Section III.

(a) Internal validation, $n=120$ , our own metric implementation
Configuration	MSE $\downarrow$	LPIPS $\downarrow$	ROI SSIM $\uparrow$	FRD $\downarrow$	AUROC ${}_{\mathrm{C}}\uparrow$	AUROC ${}_{\mathrm{T}}\uparrow$	Dice $\uparrow$	HD95 $\downarrow$
Submitted method	0.523	0.185	0.495	—	—	—	—	—
(b) Challenge debug phase, fixed 5-case cohort, official metrics — candidate selection only
Earlier ensemble	0.716	0.207	0.367	28.39	0.88	0.42	0.348	245.96
+ perceptual post-processing	0.725	0.166	0.366	29.03	0.88	0.38	0.345	246.42
Submitted method	0.569	0.159	0.360	27.88	0.84	0.62	0.347	243.42

TABLE II: Results by evaluation tier. Tiers are never compared with one another. Block (b) lists raw official metrics on a fixed five-case cohort and is candidate-selection evidence only; no rank or composite score from that phase is reported. Block (c) describes an earlier candidate, not the submitted method.

Fig. 2: Effects of lesion-localization quality and gating strategy on internal synthesis performance. Internal validation,

n=120

, our own metric implementation, a single base pipeline. Each bar is a controlled one-factor comparison against the named control; negative values indicate lower error. The ground-truth-mask oracle is not deployable. Dice values are joined from a separate segmentation evaluation, and the mean-logit Dice (

\approx 0.702

) is approximate.


Configuration	MSE $\downarrow$	LPIPS $\downarrow$	ROI SSIM ${}^{\dagger}\uparrow$	CE $\downarrow$	BG $\downarrow$	TU $\downarrow$
Unprocessed input	1.2837	0.1489	0.1952	3.0428	0.0264	3.3592
Base residual U-Net	0.5710	0.2021	0.5599	0.8615	0.1556	0.8905
Contrast-preserving refinement	0.5709	0.2084	0.5665	0.7910	0.1333	0.7940
Region-separated branch	0.5409	0.2048	0.5639	0.8222	0.1183	0.8582
Region ensemble	0.5466	0.2051	0.5638	0.8115	0.1195	0.8285

TABLE III: Internal ablations (

n=120

, our own metric implementation). Blocks come from different evaluation runs and their absolute values are not comparable across blocks; only within-block differences are meaningful. Contrast error (CE), background-over (BG) and tumour-under (TU) are in-house diagnostics, not challenge metrics. The submitted method’s headline internal values appear in Table II (a), a third evaluation run, and are not differenced against either block.

Explore similar work

Sep 27, 2026cs.CV

Anguinus Sculpturae: Compositional Synthesis of Peak-Enhancement Breast DCE-MRI Scans

Dynamic contrast-enhanced breast MRI (DCE-MRI) is rich in anatomical and perfusion information, but its reliance on gadolinium-based contrast agents raises safety concerns and adds cost. Virtual contrast enhancement, synthesizing post-contrast from pre-contrast images, is a promising alternative. We address the MAMA-SYNTH challenge task of predicting peak-enhancement breast MRI. Rather than adopting the full machinery of diffusion or flow matching, we observe that under a rectified, straight-line path the generative process collapses to a single difference prediction: the synthetic peak image is the pre-contrast image plus a predicted enhancement map, recovered in one forward pass. Around this we build Anguinus Sculpturae, a compositional pipeline in which nnU-Net segmentations of lesion, foreground and breast region guide two generators - one optimized for global fidelity, one for lesion structure through an asymmetric Tversky term routed via a frozen segmenter - composited region-wise with Gaussian-weighted blending. On the held-out Duke subset of MAMA-MIA our model achieves the best FRD and Dice among all evaluated variants, showing that single-step difference prediction with segmentation guidance suffices to recover both global fidelity and lesion structure. Code is available at https://github.com/MIC-DKFZ/AnguinusSculpturae.

Benjamin Hamm, Nico Albert Disch, Maximilian Rokuss +3

German Cancer Research Center (DKFZ) Heidelberg, Division of Medical Image Computing, Germany · Medical Faculty, Heidelberg University, Germany · Faculty of Mathematics and Computer Science, Heidelberg University, Germany +3

Image Enhancement Conditional Image Generation Magnetic Resonance Imaging Medical Image Synthesis

Ask PDF

Jul 21, 2026eess.IV

MIRAGE: Multi-scale Lesion-Informed Representation with Auxiliary Guidance for MRI Contrast Enhancement

Inferring contrast enhancement from one pre-contrast breast MRI slice is underdetermined: post-contrast appearance contains physiological information that is not uniquely encoded in baseline anatomy. Optimizing only paired pixel fidelity can suppress uncertain lesion enhancement, whereas adversarial or stochastic generative objectives can favor realistic post-contrast appearance without guaranteeing patient-specific lesion fidelity. We introduce MIRAGE, a residual 2D U-Net that combines global reconstruction and perceptual losses with three forms of lesion-aware supervision available only during training: an asymmetric penalty for missed tumor enhancement, multi-scale auxiliary tumor segmentation, and guidance through a frozen post-contrast tumor segmentation nnU-Net. We evaluate the method on 301 cases from the multi-centre MAMA-SYNTH data using eight complementary image-, region-, radiomics-, and segmentation-based metrics. MIRAGE ranks first on six metrics and markedly improves downstream lesion localization over tuned pix2pix, conditional diffusion, and latent bridge-matching baselines. The generative alternatives retain advantages in LPIPS or contrast classification, revealing a clear fidelity-utility trade-off. Leave-one-in and leave-one-out ablations show that the losses are partly redundant for lesion localization but exert distinct effects on appearance, radiomics, and boundary accuracy. These results support task-aware synthesis while also showing that its apparent optimality is conditional on the downstream models and metrics used to define utility.

Andrea Borghesi, Xin Wang, Jonas Teuwen +1

Netherlands Cancer Institute

Magnetic Resonance Imaging Medical Image Synthesis

Ask PDF

Aug 26, 2026cs.CV

MAMA-FLUX.2: Image-to-Image Synthesis of Post-Contrast Breast DCE-MRI for the MAMA-SYNTH Challenge

Dynamic contrast-enhanced breast MRI is central to cancer diagnosis and monitoring, but requires gadolinium-based contrast agents. In this work, we address pre-to-post contrast breast MRI synthesis for the MAMA-SYNTH challenge. We propose MAMA-FLUX.2, a conditional latent flow-matching approach based on FLUX.2-Klein-4B. The pre-contrast image is encoded as spatial conditioning, while the model predicts the flow field associated with the post-contrast target latent. To adapt the pretrained model efficiently, we use LoRA fine-tuning and introduce a regional training objective combining global flow matching, tumor-region supervision, and stable foreground regularization. We further investigate LoRA rank, intensity windowing, and regional loss weights on axial slices, prioritizing clinically relevant tumor-focused metrics. Our ablation study shows that moderate tumor and stable-foreground weighting improves the trade-off between image fidelity and tumor-region accuracy. The final model achieves the best overall balance with LoRA rank/

α=64/64

\mathrm{MHA}_{\max}=25

λ_{\mathrm{tumor}}=0.25

, and

λ_{\mathrm{stable}}=0.1

. These results demonstrate that compact pretrained rectified-flow transformers can be adapted for contrast-enhanced MRI synthesis using parameter-efficient fine-tuning and task-aware regional losses.

Kamil Kwarciak, Marek Wodzinski

Department of Measurement and Electronics, AGH University of Krakow, Krakow, Poland · Sano Centre for Computational Medicine, Krakow, Poland

Medical Image Synthesis Magnetic Resonance Imaging

Ask PDF